01 — Connecting the server to Antigravity

This is how I actually wired this server into Antigravity, after the first attempt failed and I went back to read the real docs instead of guessing.

Step 1 — find the real config file

I used to think the config lived in a project folder (.agents/mcp_config.json). That's not wrong, but the real global one — the one that works across every project, not just one — is at:

~/.gemini/config/mcp_config.json

On my machine this file already existed but was completely empty (0 bytes). I didn't know that until I actually checked.

Step 2 — write the server entry

The schema is only three keys: command, args, and optionally env. No cwd. I tried cwd last time and it just gets ignored — Antigravity doesn't support it, so the server never actually started right.

{
  "mcpServers": {
    "career-watch": {
      "command": "/Users/kuroko/Desktop/Nikan/career-watch-mcp/venv/bin/python",
      "args": ["/Users/kuroko/Desktop/Nikan/career-watch-mcp/server.py"]
    }
  }
}

I made server.py figure out its own folder from __file__ instead of depending on cwd, so this problem can't come back even without that key.

Step 3 — fully restart, not reload

A window reload is not enough. I have to completely quit Antigravity and open it again for it to pick up a config change.

Step 4 — verify it actually connected

There's no command like gemini mcp list here. The real way to check is typing @ in the Antigravity chat box — if the server connected, I see career-watch and its tools in that mention menu. If it's not there, the config didn't load, and I need to check the JSON for a typo before anything else.

What I'd check first next time this breaks

1. Is the JSON valid (no trailing comma, no missing quote)?
2. Are both paths in command and args absolute, and do they actually exist?
3. Did I really fully quit the app, not just reload?
4. Does @ show the tools or not — that's the real signal, not whatever the chat says it's doing.